Generative Adversarial Networks (GANs) for Robotics: Generating Realistic Data for Training, Simulation, and Bridging the Sim-to-Real Gap

Authors: Ruchik Kashyapkumar Thaker

DOI: https://doi.org/10.5281/zenodo.14001498

Short DOI: https://doi.org/g8n8x8

Country: United States

Full-text Research PDF File:   View   |   Download

Abstract: In recent years, generative adversarial networks (GANs) have emerged as a transformative technology in the field of artificial intelligence, particularly for addressing challenges in image generation, augmentation, and segmentation. This study provides an overview of the theoretical foundations and applications of GANs in robotics, with a specific focus on their role in bridging the simulation-to-reality (sim-to-real) gap. By leveraging GANs, researchers can generate highly realistic data that enhances the performance of machine learning models in tasks like object detection, crowd simulation, and autonomous navigation. Key applications include generating synthetic datasets for planetary rover localization in virtual environments and training object detectors for soccer robotics. Through a systematic review of recent publications and case studies, this paper discusses how GANs are reshaping data-driven approaches in robotics, providing insights into future research directions and the potential of GANs to advance real-world robotics applications.

Keywords: Generative Adversarial Networks (GANs), Deep Learning, Sim-to-Real Transfer, Autonomous Robotics, Computer Vision, Object Detection

Paper Id: 231434

Published On: 2022-01-16

Published In: Volume 10, Issue 1, January-February 2022

Cite This: Generative Adversarial Networks (GANs) for Robotics: Generating Realistic Data for Training, Simulation, and Bridging the Sim-to-Real Gap - Ruchik Kashyapkumar Thaker - IJIRMPS Volume 10, Issue 1, January-February 2022. DOI 10.5281/zenodo.14001498

Share this